SDataPro/maycee-retail-dataset
收藏资源简介:
Maycee Retail Dataset是一个现实的合成澳大利亚零售数据集,专为SQL学习、数据工程实践、分析工程、商业智能演示以及大型语言模型/文本到SQL评估而设计。该数据集覆盖了2017年1月1日至2019年12月31日的时间范围,包含3年的每日时间深度。具体数据包括164,968笔交易、417,907个行项目、8,762次退货,以及截至免费层窗口结束的完整当前状态维度快照。数据集由13个表格组成,包括事实表(如交易、项目、退货)和维度表(如客户、产品、商店、区域、类别、品牌、供应商、日期维度等)。这些表格模拟了零售业务中的星型模式结构。此外,数据集中还包含了故意设计的数据质量场景,例如2018年7月1日至9月30日期间的POS电话捕获中断(导致新注册客户的电话号码字段为NULL)以及2019年11月至12月交易的年终退货积压(部分退货日期在原购买窗口之后的1月至2月),这些场景用于数据质量学习和基准测试。数据集以Parquet文件格式存储,支持通过Hugging Face的datasets库加载或直接使用DuckDB查询。免费层数据基于CC BY 4.0许可发布,允许商业使用,但需注明出处。
Maycee Retail is a realistic synthetic Australian retail dataset for SQL learning, data engineering practice, analytics engineering, BI demos, and LLM/text-to-SQL evaluation. This free Hugging Face release covers 2017-01-01 to 2019-12-31: 3 years of daily temporal depth (2017-2019). It contains 164,968 transactions, 417,907 line items, 8,762 returns, and full current-state dimension snapshots as of the end of the free-tier window. The dataset includes fact tables (transactions, items, returns) and dimension tables (customers, products, stores, regions, districts, categories, brands, suppliers, date_dim), totaling 13 tables in a star-schema structure. It also incorporates deliberate data quality scenarios for learning and benchmarking, such as a POS phone capture outage from 2018-07-01 to 2018-09-30 (where customer.phone is NULL for sign-ups in this quarter) and a year-end returns backlog for Nov-Dec 2019 transactions (some returns have return_date in Jan-Feb after the original purchase window). The data is stored in Parquet files optimized for datasets.load_dataset() and direct DuckDB reads. The free tier is released under CC BY 4.0 license, allowing commercial use with attribution.




